Arun Hampapur

Digital video segmentation

The data driven, bottom up approach to video segmentation has ignored the inherent structure that exists in video. This work uses the model driven approach to digital video segmentation. Mathematical models of video based on video production techniques are formulated. These models are used to classify the edit effects used in video and film production. The classes and models are used to systematically design the feature detectors for detecting edit effects in digital video. Digital video segmentation is formulated as a feature based classification problem. Experimental results from segmenting cable television programming with cuts, fades, dissolves and page translate edits are presented.

Metadata in video databases

Video is composed of audio-visual information. Providing content based access to video data is essential for the sucessful integration of video into computers. Organizing video for content based access requires the use of video metadata. This paper explores the nature video metadata. A data model for video databases is presented based on a study of the applications of video, the nature of video retrieval requests, and the features of video. The data model is used in the architectural framework of a video database. The current state of technology in video databases is summarized and research issues are highlighted.

Indexing in video databases

A video database provides content based access to video. This is achieved by organizing or indexing video data based on some set of features. This paper defines the problem of video indexing based on video data models. The procedure required to index video data is outlined. The use of semi-automatic techniques to speed up the indexing processes are explored. These techniques use image motion features to aid in the indexing process. The techniques developed have been applied to video data from cable television feed.

Visual robot navigation using flat earth obstacle projection

Autonomous robot navigation requires the solution of several problems like obstacle avoidance and road following. Traditionally, the problem of obstacle avoidance has been solved by using range information acquired through some form of range measurements like laser range cameras or vision based range estimation. This paper presents a novel approach for obstacle avoidance which does not require true range measurement. Obstacle information is acquired using a single color camera. Detected obstacles are projected into 3D grid derived from the camera geometry and a locally flat earth model for the terrain. A weighted polar histogram technique is used to generate the vehicle steering angle based on the current grid information.<<ETX>>

Visual Motion Analysis Using Feedback

This paper presents a feedback scheme for estimating a linear global model of the image plane motion. This estimation uses a sparse image velocity field. The orientation of the individual pixel motion is treated as a vector field and a global model is estimated. The model to be estimated is chosen based on the type of motion of the robot as measured by independent sensors. The use of orientation of pixel motion makes the model independent of the structure of the scene.

Simulation for outdoor navigation

Simulation is used in robot navigation for testing control algorithms such as obstacle avoidance and path planning. Simulation is also being used for generating expectations to guide sensory processing for robots operating in the real world. In this paper, we present the tradeoffs in designing an environment model for outdoor environments. The models for outdoor environments are significantly different from indoor environment models. Outdoor environments are inherently unstructured due to changing lighting conditions, variation in the form of objects, and the dynamic nature of the environment. We present the design tradeoffs involved in building models of such environments, from the perspective of simulating passive sensors like vision. We point out that a powerful approach to outdoor navigation is to have a detailed model of the environment that can provide expectations both in terms of the spatial location of scene entities and the operators suitable for detecting these entities in an image.